Machine learning in trading: theory, models, practice and algo-trading - page 397
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
I have a shortened version, this is a set result of which I posted just above.
But then again it took me almost a day to optimize it.
The results of the importance assessment are as follows. The higher the predictor in the table, the better. OnlyVVolum6, VDel1, VVolum9, VQST10 passed the test.
In rattle we can build 6 models at once on these 4 predictors, and SVM shows accuracy of about 55% on validation and test data. Not bad.
Is it possible to get the source file? I want to compare this result with my approach.
The 55% result is very bad on these 6 predictors
Is it possible to get the source file? I want to compare this result with my approach.
The result of 55% is very bad on these 6 predictors
But then again, it took me almost 24 hours to optimize it.
Uh... no, I need something for 15 minutes :) then I'll remove half of the predictors
To be honest, I don't even know how it all works. I think it goes like this. The sample is divided into two parts, one for testing and one for training. One grid is trained on one part and tested on the other. Another on the contrary is trained on the second, and tested on the first, then the result is summed up and the overall result is calculated, so that is IMHO
This methodology applies to any machine learning.
M.B. Maxim in the subject?
Is Reshetov's classifier and that program still a single neuron, not a network? Or are they combined into a network of Reshetov neurons?
But I want to address to all. In a large set there is no data from 05.29. That is, you can teach it to the full and get the model, then you can load the model in MKUL and see how it will work for these two weeks. This will be an indicator of the effectiveness of the model. In other words, the model that will gather more with minimal drawdown and that one will win. I optimized a shortened set and the model worked like this
Let's see how your models will work in this area????
uh... no, I need something for 15 minutes :) then I'll delete half of the predictors
Well then leave del,vdel,volun,vvolum.
this methodology applies to any machine learning.
Maybe Maxim is on the subject?
I sent you a link to his website, there is a description of the model. I don't even know how to characterize it, it says Nuclear Machine + Vector Machine. It's more complicated there than in MT5 version and I want to use it to train with my opponent instead of selecting weights in optimizer, but in result I get the same weights for each of predictors.
Well now let's dream a little and imagine that we have a machine with 100 cores for optimization and we run a full dataset with 452 rows and a full set of columns and in a reasonable time the optimizer has counted everything, what would be the model????
Well first of all the input variables will be more than 10-12 and the size of the polynomial will be quite large. What does that tell you. That the model is multi-parametric, which takes into account many market factors (which is quite relevant, because it's ridiculous to predict the market based on one machine (as an example)) The length of the polynomial will indicate that the model is very flexible. As a result, such a model will work long with the proper level of quality, when the balance curve is directed upwards at an angle of 45 degrees without sharp jumps and troughs. Wouldn't that be a dream????
And as for the big set, I'll tell you that the entire June futures contract is collected there. In other words, teach the model on this data and achieve a good result in the training and testing and this model will work for the rest of its life, because it learned the whole futures contract. The next contract will be exactly the same in terms of the relationship between outputs and inputs IMHO. It's a form of grail that works with errors, but long enough. And if you train the network on the annual data, with the proper level of quality, then the market will be known to the grid. Something like this....
Well, yes. There's only the old version and the basic approach. But as practice has shown the approach with two grids increases generalization ability significantly. The result of the optimizer's work is the following file. It shows two grids and a different normalization for each grid.
So Reshetov made a normal product, you should not criticize him if honestly......
Looked at the file, there are 8 coefficients input is23 i.e. there is a neuron to work with 3 inputs. I guess, that twenty-four hours your program counts, which 3 inputs out of 100 to feed this neuron. I thought that neuron was expanded to at least 10 inputs...
getBinaryClassificator1(x0, x1, x2, x3, x4, x5, x6, x7);